AI in primary care: linking clinical and data science academics

Hello, I am Eduard, a clinical academic and GP in Liverpool with a cardiology background interest in ageing, multimorbidity and frailty. I believe that artificial intelligence (AI) has huge potential for supporting primary care. I keep waiting but have also worked with data scientists designing a number of research projects. This has changed my simplistic perception of this field and made me reflect on how to work better together.

The challenges faced by primary care continue to build up. The overwhelming amount of new knowledge about the multitude of conditions, medications, and their interactions is difficult to follow even for a motivated clinician. Even most knowledgeable clinicians cannot predict the interactions of multiple drugs and their optimal choice for a person with multiple conditions. This is even more so, when such people have been excluded from clinical trials that define the clinical recommendations. While no technology can replace human interaction between a clinician and a patient, some help may be needed to ensure the best advice and informed decisions.

Would AI help primary care in the future, and what may help primary care academics to accelerate the development? In 2018 RCGP has prepared a report ‘Artificial Intelligence and Primary Care’ to outline perspectives of AI for general practice. Still, the advance of AI into clinical practice has been slow, which may partially reflect the challenges of linking clinical and data science academics.

Why effective communication between primary care academics and data scientists is vital?  Developing AI tools requires a vast range of expertise and skills. Indeed, the diversify of niche specialities in AI and data science may be comparable to that in medicine. While every doctor has fundamental clinical training, specialised knowledge is essential for sophisticated projects. The same is valid for data science. It would be unrealistic for a clinician to know the details of all data science branches. Still, a broad understanding of its capabilities, such as deep learning, predictive analytics, and natural language processing, among many related to AI, would help build teams suitable to address the primary care problems. This would also allow the primary care academics to acquire the skillset needed to implement the new AI-driven tools. Primary care education shall prepare future clinicians for the coming changes.

Equally, the input of clinicians is essential for any AI or data science projects with ambition for future clinical use. The needs of clinicians may not be obvious to data science colleagues with the risk of new technologies providing task substitution rather than time-saving and better decisions. Close collaboration with clinical academics from the early stages of AI projects will support their future implementation.

Data shift is a major consideration for AI-based clinical tools. Data shift occurs when data used to develop AI algorithms change over time or differ substantially from the population characteristics to which the tool is applied. For example, using UK Biobank data to produce and test algorithms may not apply with the same precision to underrepresented hard to reach groups, often those from poorer socio-economic backgrounds who may be underrepresented in the training dataset. Input from clinical academics helps identify and mitigate biases and support the development of adaptive systems.

Data, data and more data, accessible, reliable and contemporary, are critical for clinical AI tool development. Accurate capture of health information and its coding is impossible without clinician input. COVID-19 pandemic has accelerated the development of integrated population health management platforms, such as CIPHA, linking primary, secondary and social care data and providing a trusted research environment.

As AI-guided tools are developed and tested, it is essential to ensure ethical and transparent data use and win patient trust. Involving clinical academics in data science projects is vital to facilitate this.

I am looking forward to working even closer with data science colleagues and using new great AI tools in my clinical practice.

Eduard Shantsila